The fabrication of trench-gated devices is common to many high performance power semiconductor devices today. Such devices include the trench-gated vertical power MOSFET (a.k.a. trench DMOS, trench FET, UMOS, etc.) and its many variants and derivatives, relying on vertical current flow through a semiconductor from a topside metal contact to a metalized backside contact whereby the potential on a trench-embedded gate electrode controls the current in said device. Such embedded trench gates actually include the gate electrode of a MOSFET whose channel is electrically induced on the sidewall of an etched silicon trench perpendicular to the wafers surface. Devices employing a trench-gated vertical channel differ dramatically from conventional MOSFETs whose gate is located above and parallel to the wafer's surface (rather than inside a trench etched into the silicon). The vertical channel allows more MOSFET gate width W to be formed in a given area of silicon real estate. The greater channel density in turn achieves a lower resistance MOSFET than devices having less gate perimeter in the same area. Such conventional trench-gated power MOSFETs are well known in the art, being manufactured since the early 1990's.
As shown in the fabrication sequence of FIGS. 1A to 1H, manufacturing a trench-gated vertical MOSFET requires a polysilicon trench gate electrode to be formed and embedded in an oxide-lined etched silicon trench. One complexity in the fabrication involves etching the trench, oxidizing it with a high quality oxide and filling the trench with a highly doped layer of polysilicon. For example, in FIG. 1A, silicon substrate having N-type epitaxial layer 10 is masked and patterned by mask layer 11 (e.g., photoresist or an etched oxide hard mask) to expose the silicon surface to etching in mask opening 12. The silicon is then etched using a reactive-ion-etch (RIE) to form a trench typically of dimensions approximately 1.5 to 2 μm in depth and 0.8 to 0.3 μm in width. The trench is then oxidized, and the oxide is removed (to eliminate crystal defects resulting from the etching process). The trench is then re-oxidized to form gate oxide 16, with a thickness between 100 Å to 1000 Å and typically around 250 Å to 500 Å. As shown in FIG. 1B, the trench is then filled with polysilicon layer 13 to a thickness sufficient to fill the trench in a void-free manner (having an embedded polysilicon portion 17A) and extend out of the trench and above the wafer's surface (so that the surface of polysilicon layer 13 becomes relatively flat).
Another challenge during wafer fabrication involves the steps of etching this gate polysilicon down into the trench and sealing it with an insulator so that it doesn't electrically short to a thick layer of source metal, said metal being formed subsequent to the trench gate and covering much of the wafer's surface. In FIG. 1C, this refill process sequence involves masking and etching back the polysilicon layer into the trench so that only the embedded portion 17A of the polysilicon gate remains in the active device array. After the polysilicon etch back, the polysilicon gate may be co-planer with the surface but typically is etched down into the trench (by less than about 0.4 μm) leaving depression 19 atop the trench. Then as shown in FIG. 1D, the silicon and polysilicon surfaces are thermally oxidized to form oxide layer 20 (having a thickness of 100 to 400 Å), followed by boron body implant 21A and 21B into all the active silicon mesa regions, typically having an implant dose of 8E12 cm−2 to 3E14 cm−2 at an energy of 80 to 120 keV to form a shallow implanted layer. This layer is subsequently diffused at a high temperature of 900° C. to 1150° C. for 3 to 15 hours to form the active channel region of the device P-type body region PB 20A and 20B, as shown in FIG. 1E. The depth of body regions 20A and 20B are typically 80 to 85% of the depth of the trench. Heavily doped N+ source regions 25A and 25B are then implanted typically with a dose of 5E15 cm−2 (or greater) of arsenic at 80 to 120 keV. Glass layer 20, e.g., comprising BPSG (borophosphosilicate glass), is then deposited and masked by a contact mask to expose a portion of the N+ silicon mesa regions 25A and 25B as shown in FIG. 1F, followed by metallization with 1 to 4-μm thick aluminum, Al—Si, or Al—Cu—Si layer 31.
Although the metal is subsequently masked and etched to separate gate and source metal connections, in the active cell array shown in FIG. 1H, the entire array is metalized by source metal layer 31. In every cell, glass capping layer 30 must prevent electrical shorts between source metal 31 and all the embedded polysilicon gates 17A, 17B, and 17C. If any one cell out of millions of cells comprising a single device becomes shorted, the entire device is ruined.
To further complicate the structure and its manufacturing, MOSFET 50 in FIG. 2A has a polysilicon gate electrode 54A and 54B that extends beyond the confines of the trenches and overlap up and onto the silicon surface, typically upon a thick field oxide region 53. The portion 55 of the polysilicon gate extending outside the trench is needed at least in one location in a device to facilitate gate contact between the embedded portions 54A and 54B of polysilicon gate and a metalized gate bonding pad (not shown). The field oxide 53 helps minimize the gate to source capacitance.
Outside the trench and atop field oxide 53, the polysilicon region 55 may extend for significant distances on and along the die's surface (i.e., outside the trench) to aid in the propagation of gate signals, i.e., to “bus” a signal, across a large area power device. Such polysilicon gate bus regions may be shorted by the topside metal along their entire length. But, since most vertical power devices employ a single layer metal process (the metal generally being several microns thick), such gate bus regions can only be metalized by interrupting the source metal. Source metal ideally, however, should cover the die's surface to its maximum possible extent (for the lowest possible resistance device). The need for a metalized gate bus therefore conflicts with the need for uninterrupted top source metal, forcing an undesirable tradeoff between fast switching speeds and the lowest possible resistance device.
The requirement for a metalized gate bus originates from the high sheet resistance of the device's embedded polysilicon gate. Even in-situ doped N+ polysilicon exhibits a high sheet resistance of 30 ohms per square, preventing signals from being bussed over intra-chip dimensions at high speeds without the assistance of regularly-distributed metalized gate-bus structures.
Also, in vertical MOSFET 50 in FIG. 2A, embedded polysilicon 54B extends out of the trench as polysilicon portion 56 crossing active area 57 before stepping up onto field oxide 53. Since the polysilicon crosses over active area 57 and thin gate oxide 62B, deep P+ region 52 must be formed beneath region 57 to prevent the thin oxide from permitting high electric fields. To form the deep P+ region 52 beneath polysilicon 56 by conventional implant and diffusion methods, the implant must precede the deposition of the polysilicon layer, preventing deep P+ region from using the same implants used to form P-type body regions 60A and 60B. Moreover, deep P+ must also precede field oxide 53 if it is to extend beneath the field oxide 52.
A difficulty with manufacturing MOSFET 50 of FIG. 2A is the non-planar topography present during critical photolithography and planarization operations. The depression of the etch-back region 66, for example, is difficult to etch uniformly. The thicker the layer of polysilicon and the thicker the field oxide, the larger the step heights present in the topography of the wafer during subsequent processing. Extreme topography can create problems during photomasking, during etching, and during planarization operations, and virtually precludes the use of chemical mechanical polishing (CMP) techniques. Large step heights may also lead to step coverage problems for depositions and conformal coatings.
Other complications in trench MOSFET manufacturing may occur at any number of steps in the fabrication sequence, and may later exacerbate issues with trench filling, planarization, and topography. For example in FIG. 2B, device 70 includes an array of trenches 73A, 73B, and 73C spaced at regular intervals as defined by mask 72A, 72B, 72C and 72D. Trench 73C represents the last trench of the regular array, whereas trench 73D represents a more distant trench, either as part of a gate bus region or a termination region distinct from the repeated array. During processing, the last trench in the array and the isolated trench regions 73C and 73D may etch differently from the regular array. Non-symmetric features 75 in trench 73C and feature 76 in trench 73D may result from optical effects during photolithography and also from micro-loading effects during etching. The misshaped trenches become very difficult to planarize using polysilicon overfill and etch-back methods since the trench opening is wider than the other trenches within the device.
Even without misshaped trenches planarization using etch-back can be challenging. FIG. 2C illustrates trench 81 (etched into epitaxial layer 80), lined with gate oxide 82 and filled with polysilicon 83. Deposited oxide 84 is then deposited and etched back to protect the embedded gate from shorting to the top metal. After deposition the surface is relatively planar compared to the thickness of the deposited glass 84. After etch-back for time t1 the depression over the trench remains constant even though the layer is now thinner. After another time t2, the depressed array nearly extends into the trench, despite the fact that the silicon mesa is not yet clear of oxide. After an additional time t3, the mesa has been cleared of the deposited glass but only a small amount of the deposited oxide remains atop the trench gate. The dielectric thickness is thinnest in the center of the trench and remains thicker adjacent to the trench sidewalls, resulting in much less planarization than might otherwise be expected using etch-back methods. This etch-back process results in all or part of top oxide 84 shown in FIG. 2D.
Later oxide 90 is deposited, masked and etched by a contact mask and subsequently flowed at a high temperature to round its shape. After completing contact mask metallization 87 completes the structure.
As shown in the drawing of FIG. 2D, several problem areas may occur in such a device. Curvature at the bottom of the trench in region 100, for example, may lead to gate oxide thinning and low gate breakdown voltages. Using only aluminum as the metallization, metal spike 102 may alloy through the N+ junction 86 and producing undesirable junction leakage. Furthermore, polysilicon gate 83 must vertically overlap the bottom of N+ source 86 in region 102 or the MOSFET will not operate properly (causing high on-resistance and possible loss of functionality).
In FIG. 2E, a TiN layer 88 is deposited after the contact mask but prior to deposition of metal layer 87. The titanium nitride layer blocks the aluminum metallization from alloying through the junction thereby eliminating metal spiking. If the step height of the oxide cap 90 atop the trench is too steep, the TiN layer 88 may crack, and spike 104 may result. In each case shown, poor planarization leads to defects, poor uniformity, and yield loss.
The problem of topography complicating planarization is further illustrated in FIG. 3A, which shows a cross section 120 of an active trench array of trenches 123A, 123B, and 123C along with gate bus area 124, just after polysilicon deposition and partial etch-back producing a polysilicon gate bus 131G of thickness x atop oxide 125. If thickness x is minimized to prevent step height related step coverage problems later, the depth of the surface depressions 135A, 135B, and 135C at the top of the active polysilicon trench gates will be extreme and may be difficult to fill later.
Cross section 160 shown in FIG. 3B illustrates the contrasting case where the polysilicon layer of thickness x is sufficiently thick to exhibit a more planar surface over the trenches, whereby depressions 162A, 162B, and 162C are at a minimum. FIG. 3C illustrates a benefit of using very thick polysilicon in device 160 is the relatively good planarization results and uniform etch-back of embedded polysilicon regions 166A and 166B. Thick polysilicon, e.g., over 1 to 1.5 μm thick, requires long processing time, adversely adding to wafer costs. More significantly, another problem with such a thick polysilicon layer is also revealed in FIG. 3C, where polysilicon gate bus 165 comprising embedded portion 166C and surface portion 167 produces two extreme steps in the wafer's topography: step 180A over active areas and step 180B on top of field oxide 181. Either location may later cause problems with glass depositions, TiN metallization, or metal breakage.
FIGS. 4A to 4J illustrate the fabrication sequence of a trench-gated MOSFET with a self-aligned contact as disclosed in U.S. Pat. No. 6,413,822, entitled “Super-Self-Aligned Fabrication Process Of Trench-Gate DMOS With Overlying Device Layer”, which is hereby incorporated by reference in its entirety. The self-aligned contact method allows the contact feature to be opened over the entire active device array device without leading to gate-to-source shorts. This feature allows improved body contact and/or denser trench array cell features and smaller devices. In the fabrication sequence, FIG. 4A illustrates MOSFET trench cell region 200 in which a composite hard mask comprising oxide 205, silicon nitride 206, and top oxide 207 is patterned by photolithographic means and then used to act as a hard mask for silicon trench etching to produce trench feature 201. Gate oxide 204 is then formed after appropriate sacrificial oxidations. Unlike in other trench fabrication sequences, in the sequence beginning in FIG. 4A, the hard mask remains on the silicon during subsequent processing. In FIG. 4B, thick bottom oxide 211 is deposited directionally using CVD. Additional oxide 210 then forms atop the hard mask during said directional deposition with little or no deposition on the trench sidewall.
In FIG. 4C, polysilicon 215 is deposited to fill the trench to a thickness that overflows the trench and deposits above the silicon surface. The polysilicon overflow is then etched back to form embedded polysilicon 215 as shown in FIG. 4D, top oxides 210 and 207 are removed as shown in FIG. 4E, and a P-type body region is implanted through nitride 206 into the silicon mesa regions interposing the trenches as shown in FIG. 4F. The body may be implanted at a shallow depth with a low energy ion implant (i.e., with under 150-keV boron ions) then diffused to a depth of substantially 85% of the trench depth, or alternatively, the body may be entirely implanted using a chain-implant of multiple boron implantations of differing energies (ranging from up to 2 MeV for the deepest portions to 150 keV for the shallowest portions) all through a common mask opening.
After the body implant, in steps not shown, a second polysilicon layer is deposited, masked, and etched back to leave polysilicon only in the surface gate bus regions and where the trench polysilicon connects to the polysilicon gate bus. After the second polysilicon etch-back, thermal oxide 226A is grown followed by glass or BPSG deposition of dielectric 225A as shown in FIG. 4G. In FIG. 4H, dielectric 225A is etched back below the surface of original nitride hard mask 206. Thereafter, nitride hard mask 206 is removed to result in the structure of FIG. 4I where only a small portion of glass 225A remains above the trench. In device 240 of FIG. 2J which represents an array of active trench devices 220, the contact mask (needed to contact gate bus regions not shown in the cross section) opens all the active areas to the contact etch but removes minimal amounts of oxide so as to avoid shorts to embedded polysilicon 215. Thereafter, TiN 230 and top metal 231 are deposited, patterned, and etched.
In FIG. 5A, the gate bus region of the same device is illustrated after the second polysilicon deposition and masked etch back. Polysilicon gate bus 226B remains on the surface of the device while only a small portion of the second polysilicon 226A may survive etch back inside the trench itself. The final polysilicon etch back uniformity is critical since any polysilicon protruding from a trench will short the device's gate to its source metal. Only etch back methods may be employed to planarize the polysilicon since the surface of the device is not planar, having various steps, and topographic features above and below the wafer's surface. Such topography prevents the use of chemical mechanical polishing (CMP) techniques, since mechanical planarization of the trench polysilicon would remove the gate bus entirely from the device's surface.
With such uneven topography, the need for glass 225A (shown after its deposition in FIG. 5B) becomes tantamount for preventing step coverage problems. The glass 225A is flowed as illustrated in FIG. 5C using a short yet relatively high-temperature furnace operation (typically 900° C. for 15 minutes) and patterned as shown in FIG. 5D according to a contact mask. FIG. 5D illustrates the contact to the gate bus (above the silicon surface), and the large self-aligned contact open across the top of all the active trench cells. The contact masking operation is complicated by the two distinct heights of the active array and the gate bus regions, but since no critical feature is present within the cell array, the contact etch operation is possible provided across-the-wafer uniformity is not at issue.
In FIG. 5E, TiN barrier metal is deposited to a thickness sufficient to prevent metal spiking in the contact windows and to cover the topographical steps, but thin enough not to crack from film stress. In FIG. 5F the thick metal, typically Al—Cu or Al—Cu—Si is deposited to a thickness of about 3 μm, masked and etched. Metal 270A represents the source metal, where 270B is the gate contact or gate bus.
FIG. 6A illustrates a plan view of a trench gated power MOSFET having a source metal electrode 283, a metalized gate pad 280, and a gate bus 281. A gap 282 separates the two metals and prevents shorting. The drain contact is made to the wafer's backside. FIG. 6B illustrates another trench MOSFET, having three separate source metals 285A, 285B and 285C, broken into islands by gate metal comprising metal gate ring 286A and busses 286B and 286C connected to metalized bonding pad 284. In such a device the separate source regions are not electrically shorted until wire bonding and packaging.
Device areas can be substantial with dimensions spanning several millimeters across a die. Uniformity across a die and across a wafer can therefore be problematic in achieving highly manufacturable products. Typical uniformity issues can manifest themselves as irregular and random metal voids such as 300A and 300B in contact openings of a device 290 of FIG. 7A. Even in a MOSFET 310 having self-aligned contacts as shown in FIG. 7B, polysilicon deposition, oxidation and etch back can lead to “horns” and other irregular etch features like the top of polysilicon gate 314B and voids as shown in gate region 314C.
Topography can also lead to steps coverage and depth of focus problems for photolithography, including potential gate shorting in region 350, TiN cracking (region 351 and 353), and metal thinning and reentrant angles over steep steps 352 and 354 as shown in FIG. 7C. All these failures may compromise the performance, yield, or reliability of a trench MOSFET. The problems are greatly exaggerated by the non-planar surface of the device, requiring photolithographic-masking, etching, etch backs, planarization, and depositions to behave similarly (or identically) on different heights within a single device.
Not only does the extreme topography of today's vertical trench power MOSFET processes limit the device's cell density and performance, its high intrinsic gate resistance limits its switching speed, mandating the need for additional metalized gate bus regions. The schematic of FIG. 8A, representing an array of MOSFETs can be used to examine how distributed gate resistance affects the switching speed of large area power MOSFETs. The lumped element of a single three-terminal power MOSFET 360 comprises an array of identical MOSFETs 362, whereby transistors M1 through Mn are connected with their source-drain terminals in a parallel network configuration sharing a common source electrode 366 and a common drain electrode 368. Although the devices also share a common gate connection 365, the gate resistance is not constant for the various devices. Devices near a gate bus, e.g. M1, conduct the gate signal through a small series gate resistance of rg1 where device M2 exhibits a higher series gate resistance of (rg1+rg2). With a higher gate resistance, the more remote device switches slower than those devices located nearest the gate bus. Cell Mn, farthest from the gate bus having a series gate resistance of (rg1+rg2+ . . . +rgn) can suffer serious gate propagation delays, even switching ten times slower than cells near the metalized gate bus. So the farthest device from the metalized gate bus is slowest in both turn-on and turn-off transients, being the last device to turn on during a turn-on transient and the last device to turn off during a turn-off transient. As illustrated in plot 370 of FIG. 8B, driving gate input signal 371 from “on” to “off”, and then to “on” again causes drain current through a resistive load to roughly “follow” the input waveform after some time delay. In ideal case 373, the drain current changes quickly after the input transition. In drain current waveform 372, which includes distributed gate resistance, the device experiences a turn-off delay td(off) followed by a slow fall in current ID for duration tfall, the fall time of the device. Turn on has a similar response with a delay Td(on) followed by a slow ramp up in current for duration trise, the rise time of the device. During slow transients, the simultaneous presence of both voltage across the device and current through the device greatly increases power losses in the device and lowers the efficiency of using such a switch in many applications. The effective gate resistance can be reduced by including gate busses regularly and more frequently throughout the device, but only by sacrificing area from active device arrays for gate bussing. More gate bussing and less active cells increases the switch's on-resistance or die size, and therefore cost. Moreover the gate bus regions exacerbate the non-planar topography issues that further limit device density and performance improvements.
What is needed is a vertical trench-gated power MOSFET capable of integrating large arrays of active vertical MOSFETs at high densities with integral gate bus and gate contact structures in an area efficient device having a relatively flat or planar surface topography. Ideally, the device should offer the lowest possible series gate resistance for fast switching capability, and exhibit low drain-to-source area-specific on-resistance. Finally, the fabrication of such a device should accommodate processes for achieving better film planarization uniformity than that of standard etch back methods.